QuEra's Yuval Boger Expects Materials Before Drug Discovery on Fault-Tolerant Quantum Computers: What the Harvard, Anthropic and QuEra Preprint arXiv 2610.22367 Cut From Chemistry Circuits
What Gu, Kemeny, Yelin and Lukin reported in arXiv 2610.22367 on 15 October 2026
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
In an interview that R&D World published on 17 November 2026, Yuval Boger, chief commercial officer of the Boston neutral-atom company QuEra Computing, said he does not expect drug discovery to be the first major payoff of fault-tolerant quantum computers. He expects strongly correlated materials to come first, with molecular chemistry, including metalloprotein active sites, following on QuEra's next-generation system. Much of the interview concerned a preprint in which a large language model designed quantum circuits for chemistry and other tasks that need fewer gates and fewer qubits than the designs the authors compared them against.
For pharmaceutical and hospital research teams the two pieces belong together. The preprint lowers the estimated cost of several building blocks that a quantum chemistry calculation needs, and the interview sets out which problems one of the companies building the hardware expects those savings to reach first, and on what timetable. Neither involves a run on quantum hardware. The preprint's circuits are written for fault-tolerant hardware, and the interview's timeline concerns machines that QuEra plans to deliver from 2028.
The Harvard, Anthropic and QuEra preprint on arXiv, posted on 15 October 2026, lists Andi Gu of Harvard's Department of Physics and the Harvard Quantum Initiative, who is also part of the Anthropic STEM Fellows Program; Alek H. Kemeny of Anthropic; Susanne F. Yelin of Harvard; and Mikhail D. Lukin of Harvard and QuEra Computing. It has not been peer reviewed. The authors built an evolutionary search in which a language model, Anthropic's Claude Mythos 5 running inside the open-source OpenEvolve framework, proposes programs that generate circuits, and an independent evaluator checks each candidate for correctness and scores its cost. For each of eight tasks the model receives a fixed, hand-written specification and a shared circuit-building interface. The tasks span numerical function evaluation, chemistry, optimization, quantum simulation and number theory.
The cost measure is the one that dominates estimates for error-corrected machines: the number of Toffoli gates, which are expensive to carry out fault-tolerantly, and the number of qubits a circuit occupies. In seven of the eight tasks the discovered circuits reduced both counts below the comparison baselines. For one decoding task the authors note that no earlier reversible compilation existed in the literature, so the baseline there was a naive reversible implementation. In the eighth, modular inversion over a prime field, a step in cryptographic arithmetic, the search produced a new reversible architecture that substantially improves the trade-off between Toffoli gates and qubits.